resolve does not preserve `require.extensions` list and assumes only ['.js']

[snuggs]

Greetings!

Having a situation where we are using tape which does not pass extensions to resolve. Have opened an issue here tape-testing/tape#395

That being said our situation is we use the .es ECMAScript extension with the following:

require.extensions ['.es'] = require.extensions ['.js']

Although we can achieve our needs with the previous tape issue/pr I feel it behoves resolve to use the default require.extensions list provided by node instead of only assuming ['.js'] here

resolve/lib/sync.js

Line 23 in 5f6e682

var extensions = opts.extensions || ['.js'];

.

An easy enhancement would be to convert:

var extensions = opts.extensions || ['.js'];

to

var extensions = opts.extensions
  || Object.keys (require.extensions)

Please advise and will have a PR & tests up ASAP as we need this or have to call off using tape and resolve. Can also enhance documentation which should close #134 which @ljharb, and @call-a3 are working on as well.

Thanks in advance! /cc @brandondees

[ljharb]

This would be a breaking change; require.extensions starts with .js, .json, and .node.

You can certainly use resolve with require.extensions via the extensions option, however.

[TehShrike]

If this package's goal is to implement the node.js require.resolve() algorithm, isn't this something that should be fixed?

Using a truncated default extensions list makes this library default to non-node-like behavior.

Packages like got that require('./package') will break when bundled by bundlers that use node-resolve (which is all bundlers that I'm aware of).

[bmeck]

It should be noted that both NODE_PATH and require.extensions have been deprecated for years. They also do not affect the ESM resolution algorithm. I would recommend moving off of relying on their usage.

I think it's good to not rely too heavily on global runtime vars to configure default values. The whole point of this package instead of using require.resolve() is that you can configure its behavior without mutating the global runtime environment variables.

[TehShrike]

That makes sense to me. Changing the hardcoded default to match the extensions given in the docs seems preferable.

[snuggs]

@TehShrike hence https://github.com/browserify/resolve/pull/138/files

This way even if someone is on some ancient deprecated api that's technically superfluous to the intent of what the library is assumed to accomplish. Feels quite awkward one would have to pass in the standard require.extensions to override the default require resolve "extensions". P.O.L.A. is real.

[TehShrike]

Interested parties should go review #145.

[ljharb]

Closing this in favor of #166, to centralize further discussion.